The Election Begins Before the Ballot. Who Is Counting the Invisible Votes?
Allegations of unequal treatment by digital platforms expose a gap in the protection of Brazilian democracy: we can audit voting results, but we still do not know how to adequately audit the private power that determines the circulation of information during a campaign.
Fernando Fagundes Ferreira ·
A democracy incapable of protecting the conditions in which voters form their will risks extraordinarily perfecting its counting mechanisms while neglecting the mechanisms of influence that precede the vote. Brazil has developed a sophisticated system to protect the integrity of its voting machines, but it still lacks equally effective instruments to oversee digital platforms' power over the circulation of political information. Allegations of unequal treatment among candidates in the 2026 elections expose this contradiction: what good is ensuring that every vote is correctly counted if economic actors can covertly interfere with the conditions in which those votes are won?
An election can be technically sound in its vote count and still be subject to illegitimate interference during the formation of the popular will.
It is precisely this contradiction that the events of October 2026 compel Brazil to confront.
The allegations brought by Lula's campaign against Meta, involving supposed restrictions on the reach of political content and unequal treatment of candidates by the company's artificial intelligence, have placed a particularly uncomfortable question before the Superior Electoral Court (TSE): who oversees the economic actors that have come to control a significant share of campaign visibility?
And, above all, what happens to an election's legitimacy if it is proven that these actors systematically interfered with the conditions of competition?
Voters see what the platform allows to circulate
Social networks are not simple display windows in which candidates place their campaign materials. They are systems that actively distribute information, capable of deciding, through automated recommendations, rankings, filters and moderation mechanisms, which content will reach large audiences and which will remain practically invisible.
Following a particular candidate does not necessarily mean receiving their posts. Producing relevant information likewise does not guarantee that it will reach its intended audience.
Between the sender and the citizen lies a private infrastructure that selects, ranks, recommends, restricts and distributes content on an industrial scale.
This does not mean that every difference in reach constitutes manipulation or electoral interference. Algorithms respond to multiple variables, including preferences, patterns of interaction and content characteristics. The issue arises when these decisions produce treatment incompatible with platforms' legal obligations or when apparently neutral criteria conceal systematic favoritism.
At that point, digital distribution ceases to be merely a commercial matter and becomes part of the problem of democratic integrity.
A company need not alter a single vote recorded in a voting machine to improperly influence the conditions of an election. Depending on the conduct, asymmetrically controlling the circulation of information that precedes voters' decisions may be enough.
Meta and the test of equality
On October 14, CNN Brasil reported on tests conducted with Meta AI, an artificial intelligence tool integrated into the company's digital ecosystem.
Requests to generate election-related images of Lula were refused, while equivalent prompts involving Flávio Bolsonaro produced the requested materials. The report reproduced examples submitted by users and conducted its own tests, identifying different responses to similar requests.
Lula's campaign brought the matter before the TSE, alleging unequal treatment, algorithmic selectivity and harm to its conditions for electoral communication.
Meta replied that it had established temporary restrictions on recommending candidates and generating their images, adding that it was improving the filters applied during the election period.
This explanation must be examined technically, not simply accepted or rejected for political convenience.
An inconsistency in an artificial intelligence system's filters does not, on its own, prove a deliberate operation to favor a candidate. It may result from implementation failures, incomplete adjustments or other factors that only an independent investigation can clarify.
But the absence of proof of intent does not make the observed effect irrelevant.
Platforms' responsibility cannot depend exclusively on what they declare they intend to do. It is necessary to verify what they actually do, how their systems operate, which criteria they apply and what consequences they produce.
The question is elementary: if two candidates are subject to the same rules, why did equivalent prompts produce different results?
And the answer cannot depend solely on the word of the company responsible for the system under scrutiny.
The dispute over reach has also reached the TSE
The problem is not limited to generative artificial intelligence.
Decisions about algorithmic recommendation, moderation of WhatsApp channels and classification of election-related content have come to occupy a central position in the presidential contest.
On October 15, the TSE unanimously struck down the requirement to wait 48 hours before reactivating WhatsApp channels associated with political campaigns. The controversy involved, among others, channels called Exército do Lula and Tropa do Flávio, subject to disputes over their formal links to the respective candidacies.
At the same time, decisions concerning the recommendation of profiles of politicians elected in the first round raised questions about possible asymmetric effects on the circulation of content favorable to candidates remaining in the presidential race.
Each decision has its own grounds and circumstances. It would not be correct to automatically treat them as part of a single coordinated operation.
Taken together, however, they reveal a structural transformation: Electoral Justice no longer administers only campaign advertising, candidate registration and voting procedures. It must also confront a communications architecture that directly interferes with the conditions under which candidates gain access to the public sphere.
The problem is that the speed of platforms and the complexity of algorithmic systems are incompatible with exclusively reactive oversight mechanisms.
A judicial decision issued today may correct an irregularity, but it does not necessarily recover the audience that a particular piece of content failed to reach yesterday.
Time lost in an election campaign is not a renewable resource.
The fraud that need not touch the voting machine
A fundamental distinction must be drawn between two dimensions of electoral integrity.
The first concerns the integrity of voting and vote counting: ensuring that every vote is recorded, preserved and correctly counted.
The second concerns the integrity of the conditions of competition: ensuring that no participant obtains unlawful advantages through economic abuse, information manipulation or improper use of communications instruments.
The first does not replace the second.
Brazil has reasons to protect and defend its electronic voting system against unfounded allegations of fraud. But it would be a mistake to turn that defense into a limited conception of democracy, according to which an election is necessarily legitimate because its votes were correctly counted.
Article 22 of Complementary Law No. 64/1990 provides for investigating abuse of economic power and improper use of social communications media. Its application considers the gravity of the circumstances, not merely a mathematical demonstration that a particular unlawful act changed the final result.
This does not mean that any moderation failure, difference in reach or inconsistent behavior by an artificial intelligence system warrants annulment of an election.
It means that the law already recognizes an essential issue: the legitimacy of an election also depends on the conditions under which the contest takes place.
Digital platforms have added a technological dimension to this legal problem.
Potential electoral interference can occur without tampering with voting machines, falsifying vote reports or directly intervening in the aggregation of results.
It can occur through the selective distribution of information, improper amplification of content, discriminatory application of rules or the use of economic structures to produce illegitimate communications advantages.
These are hypotheses that need to be investigated, not conclusions that can be presumed.
Should the first round be annulled?
The question cannot be dismissed simply because its possible consequences would be politically inconvenient. Nor can it be answered affirmatively on the basis of allegations that have not yet been sufficiently substantiated.
Annulment of an election is an exceptional measure, subject to rigorous legal, procedural and evidentiary requirements. It is not equivalent to a candidacy's dissatisfaction with its electoral performance.
However, it would also be unacceptable to establish in advance that no interference by digital platforms could ever be serious enough to compromise an election's legitimacy.
The institutionally responsible path begins with an independent, comprehensive and technically qualified investigation.
It is necessary to determine whether irregularities occurred, identify their nature and extent, establish responsibility and examine whether they produced unlawful advantages sufficiently serious to justify electoral consequences.
The investigation must consider the period before the first round and, with particular urgency, the current conditions of the presidential contest.
This is not about creating a retrospective narrative to justify candidates' performance. It is about preventing potential irregularities from becoming normalized simply because they occurred before the voting machines opened.
If the rules of an election were gravely violated, the solution cannot consist of acknowledging the problem and automatically preserving all its effects.
On the other hand, no election should be annulled on the basis of conjecture, political inferences or accusations lacking technical substantiation.
Democratic commitment demands both: the willingness to investigate and the responsibility to decide.
Five measures the TSE should prioritize
Protecting electoral integrity must extend to the digital mechanisms that shape the circulation of political information. This requires concrete measures compatible with the law, due process and the protection of fundamental rights.
1. Immediate preservation of relevant data. Order, within legal limits, the retention of technical records necessary to investigate potential irregularities in the distribution, recommendation and moderation of election-related content.
2. Independent technical audit. Examine the mechanisms under scrutiny, the criteria actually applied and the results produced, with the participation of independent experts and proportionate protection of trade secrets, personal data and protected information.
3. Verification of unequal treatment. Compare equivalent situations among competing candidacies, distinguishing technical failures, regular application of moderation policies and possible discriminatory or unlawful practices.
4. Transparency about decisions affecting electoral reach. Require reasoning and sufficient information to assess measures capable of significantly restricting the circulation of political expression, without presuming that all moderation constitutes censorship.
5. Proportionate and timely accountability. Apply legally available measures when irregularities are established, considering their gravity, extent, responsibility and effects on the contest.
This is not a proposal for the state to assume political control of social networks or determine which opinions deserve to circulate. That would merely replace one private concentration of power with another, state-controlled one, equally incompatible with a democratic society.
The aim must be to ensure that neither of these structures holds arbitrary power over the conditions of political participation.
The new frontier of lawfare
Contemporary lawfare is not limited to the strategic use of judicial proceedings against adversaries. Its investigation must also consider the ways in which legal, economic, technological and communications structures may converge to alter the conditions of a political contest.
This possibility does not turn every unfavorable decision, every business practice or every regulatory intervention into lawfare. It requires identification of the concrete mechanisms of instrumentalization and their institutional effects.
But it would be equally naive to imagine that political power remains confined to the institutions traditionally recognized as part of the state.
Companies that administer communications infrastructures used by millions of citizens possess a potentially significant capacity to interfere in the public sphere. When that power operates without sufficient transparency and accountability, popular sovereignty itself is exposed to a structural dependency.
The question, therefore, is not merely which candidacy was favored or harmed by a particular episode.
It is whether Brazilian democracy has instruments capable of identifying, investigating and confronting unlawful interference produced by organizations that do not run in elections but control decisive means for the circulation of campaigns.
Popular sovereignty cannot end where a digital platform's proprietary code begins.
Who counts the invisible votes?
We are accustomed to following election counts through numbers. Valid votes, invalid votes, blank votes, percentages, abstentions and consolidated results.
But there is a dimension prior to the count that remains largely invisible.
How many people failed to receive a particular piece of information because of an improper restriction? How many were reached by artificially favored content? How many political expressions had their circulation reduced by inconsistent automated decisions? How many of these differences arose from platforms' regular mechanisms, and how many could constitute unlawful interference?
The aim is not to turn these questions into an imaginary accounting of votes that we will never be able to attribute precisely to a specific post or algorithm.
It is to recognize that the formation of political will has come to depend, in part, on private infrastructures whose operation must be subject to scrutiny commensurate with their democratic importance.
The debate on electoral integrity cannot begin when voting opens and end with the announcement of the results.
The election begins long before the voting machine.
It begins with access to information, the ability to challenge power, the circulation of different political interpretations and legal equality in the conditions of competition.
A democracy incapable of protecting these dimensions risks extraordinarily perfecting its counting mechanisms while neglecting the mechanisms of influence that precede the vote.
Democracy cannot limit itself to ensuring that all votes are correctly counted. It must also ensure that no economic actor holds the covert power to decide which candidacies will have the conditions to compete for those votes.
Auditing the voting machine is not enough.
We must illuminate the power that operates before it.
Lawfare is for everyone. Justice is not.
Sources and references
CNN Brasil (October 14, 2026). Entenda a acusação da campanha de Lula contra a Meta.
https://www.cnnbrasil.com.br/eleicoes/entenda-a-acusacao-da-campanha-de-lula-contra-a-meta/
Agência Brasil (October 15, 2026). Campanha de Lula denunciará Meta na OEA por interferir nas eleições.
https://agenciabrasil.ebc.com.br/radioagencia-nacional/politica/audio/2026-10/campanha-de-lula-denunciara-meta-na-oea-por-interferir-nas-eleicoes
Folha de S.Paulo (October 15, 2026). TSE derruba regra que exigia espera de 48 horas para reativar canais de WhatsApp.
Complementary Law No. 64/1990, Article 22. Electoral judicial investigation and abuse of economic power or improper use of social communications media.
Editorial note: This article examines allegations and events known as of October 15, 2026. It does not assign definitive responsibility to platforms, candidates or judicial authorities for facts still subject to investigation.